RESEARCH × OPEN SOURCE
AI/ML PhD student
Université de Montréal & Mila
Rule alignment for language models.
I'm Wenhao XU. I study whether language models actually use the rules we give them.
I also build open-source skills for inspecting, replaying, and challenging the output of agents.
Ask the AI guide about my work, or try research, skills, or help.
Questions and the last two exchanges are sent to Cloudflare Workers AI. This site does not save chat text; context stays in this tab until you clear it or reload. Daily salted network hashes and request counts expire at 00:00 UTC. The shared free allowance is limited. The guide uses curated public facts, may make mistakes, and does not speak for Wenhao. Cloudflare data usage ↗
Tab to complete ↑ ↓ historyNo shell. Just exploration.
Follow a path, then remove the rule.
An explicit instruction. Does it change the decision, or is the model following a shortcut?
Rule present → decision A
Illustrative circuit · not a trained model or research result.